oss_13_http_exporter_retry_after_backoff_cap

[ocelotl]

Closes #18

OTLP HTTP exporter retry conformance (Linear OSS-13, findings X2/X3): make HTTP 429 retryable, parse and honor Retry-After (delay-seconds + HTTP-date), and clamp exponential backoff to a 32s ceiling across the three HTTP exporters and the gRPC exporter.

Validation: scope contained to the OTLP http+grpc exporter packages + changelog; HTTP suite 73 passed, gRPC suite 67 passed.

Linear issue: https://linear.app/dash0/issue/OSS-13/x2-x3-http-exporter-honor-429retry-after-add-max-backoff-cap

[ocelotl]

🔒 Internal (dash0) — not for upstream.

• Linear: https://linear.app/dash0/issue/OSS-13/x2-x3-http-exporter-honor-429retry-after-add-max-backoff-cap (findings X2/X3)
• Closes: oss_13_http_exporter_retry_after_backoff_cap #18 · Branch: oss_13_http_exporter_retry_after_backoff_cap · Commit: c29c774bc
• Validation: HTTP 73 passed, gRPC 67 passed; scope otlp http+grpc + changelog.

[ocelotl]

📣 Public-facing draft — to be used for the upstream PR in open-telemetry/opentelemetry-python. No internal references; copy verbatim.

Title: Retry OTLP HTTP exports on 429, honor Retry-After, and cap backoff

What changed

• 429 is now retryable (HTTP): _is_retryable() in opentelemetry/exporter/otlp/proto/http/_common/__init__.py returns True for 429, in addition to 408 and 5xx.
• Retry-After is honored (HTTP): a new _parse_retry_after_header() parses both supported forms — delay-seconds (Retry-After: 120) and HTTP-date (Retry-After: Wed, 21 Oct 2015 07:28:00 GMT). A past HTTP-date yields 0; a missing/malformed header yields None. When present, its value overrides the computed backoff as the next-retry delay in the trace, metric, and log HTTP exporters.
• Backoff is capped: a _MAX_BACKOFF (32s) ceiling is applied via min(2**retry_num * random.uniform(0.8, 1.2), _MAX_BACKOFF) in the three HTTP exporters and in the gRPC exporter.

Why

The OTLP exporter specification requires retrying on 429 and honoring Retry-After, and both Go and Java cap their backoff. This brings the Python HTTP exporters into conformance and bounds retry delays.

Spec compliance

• 429 retryable: matches the spec and the existing gRPC behavior.
• Retry-After covers both RFC 9110 forms, matching Go/Java.
• Backoff clamp matches Go/Java.

How tested

New unit tests: 429 triggers a retry; Retry-After (seconds and HTTP-date) is used as the delay; backoff is clamped at the cap; plus direct tests of _is_retryable and _parse_retry_after_header (missing / seconds / HTTP-date / past-date / invalid), across the trace, metric, and log HTTP test modules. Full OTLP HTTP suite passes (73); full gRPC suite passes (67).